Hey Todd, GH, and all,
I would like to make this run the topic of an article in
NAP. I'd like to address some of your questions there, as
well as a few others. Don't worry, Todd, you'll look like
the great driver that you are when I'm done! ;-) Only
7/10ths behind Hollywood Bob is a great job in most anyone's
book!
Much of this will be in the article which will probably be
in the June issue of North American Pylon. I'll spend a
little more time editing it for cleanliness, but this will
basically be the June article.
Specifically I'd like to get GH's and Todd's permission to
use snippets from their e-mails within the article, and
their names. If that's okay, guys, just drop me a note
saying so. My deadline is tomorrow (Monday) I believe.
The ratings systems we use with GEEZ are not perfect, of
course, but the correlation to quality driving is extremely
high. In cases where you compare the same car, same course,
and same driver, GEEZ will regularly correctly order about
85-90% of the runs, even when the differences in time are as
low as 1 or 2 tenths of a second.
When you begin to use two different drivers, in two
different, but similar cars, the correlation diminishes
somewhat, but is still generally in the 75-85% range or so.
Again, we are actually pretty proud of this high correlation
rate. Keep in mind, GEEZ doesn't really care about your
time in making these determinations, rather, it's just
looking at the various definitions of "quality" that we've
given it, vis-a-vis smoothness, aggressiveness, and usage.
The fact that we have such a high correlation is somewhat
still an amazement, even to me.
By setting up fancy little algorithms that we think measure
different aspects of driving skill at a very minute level,
the software scores every little move you make, most of them
on two different scales. Then when it's all done, it
produces a completely objective number. The predictive
quality of that number to the time generated is at times
really uncanny.
However, it seems like you are wondering if the ratings are
too high for your particular case, or if the smoothness is
really that important to you, or both. Well, probably "yes"
and "yes".
To cover a couple of quibbles about how GEEZ computes some
of these ratings, and why, let's first talk about the issue
of Peaks as they are used as benchmarks for the various
ratings.
First, it's not necessarily true that the better driver
pulls higher peak g levels. In fact, at least as often as
not, the opposite is true, especially when using Absolute
Peaks. The less skilled driver hits the car's peaks all
right, in fact, he often attempts to go right past them.
This of course doesn't make us fast. The more skilled
driver may hit the same levels, or sometimes even lower.
The difference is, he stays there longer.
So while the peaks hit can and do vary from driver to driver
in the same car, and even from run to run with the same
driver, the tendency is for the level of the peaks to be
less tied to the driver, and more to the car. It's not
perfect, but that tendency is noticeable.
Now when comparing two great drivers like Bob Tunnell and
Todd Green, I'd be surprised to see any substantial
differences in peaks between the two of them. Both are
certainly accomplished enough to find the cars limits, and
the car isn't likely to have "hidden talent" that only Bob
Tunnell can find. The car's peaks tend to be the car's
peaks.
So the effect mentioned by GH and Todd about the peaks
changing around is certainly possible, and always occurs to
some level, it's not where I'd look for the possible time
improvement.
Most of us have our program defaults set to measure us
against "Sustained Peaks" with a Sustain Duration of 0.5
seconds. This is the factory default. The ratings are
directly related to these peaks as determined by the 1/2
second sustain. As you move the sustain value shorter, your
ratings will generally drop, because you have a higher
target to hit in order to score well.
So if you find that your ratings are just too high (a nice
problem to have, but perhaps a problem nonetheless), I would
suggest shortening your sustain duration. This will help to
yield a different picture of your driving ability. Now
remember when you do this that any driver whom you are
comparing should be judged by the same yardstick.
Going to Setup|Program Options|Data and setting the Sustain
Duration to 0.3 seconds, for instance, changes the ratings
to: Overall 99, Usage 99, Smoothness 94, and Aggressiveness
96. This begins to show that if we get pickier about things
that the Smoothness actually gets better, and the
Aggressiveness gets worse. This is because we now compare
to higher limits, and the build-up and drop-off of lateral
g's used in smoothness become less of an issue, thus helping
smoothness, while the aggressiveness suffers because it's
harder to "snap" to the maximums as quickly.
Eliminating the Sustain entirely, by selecting "100% Levels
are based on Absolute Peaks" (or by setting the Sustain
Duration to 0.1 second) gives even more detail. The ratings
are now: Overall 95.60%, Usage 97%, Smoothness 94%,
Aggressiveness 92%.
What a change this is! By demanding that we use the
absolute peaks that the car can reach as our upper limit, we
now see that rather than worrying about being too
aggressive, as we originally thought with the program
defaults, we might in fact want to try to increase
aggressiveness even a little bit more. One word of caution
though--if we try to increase aggressiveness and sacrifice
smoothness doing it, we are going to begin losing points in
Usage, and quick!
How is it going to be possible to get more aggressive? The
answer is likely to increase smoothness. Let me show a
couple of places in particular that I think increasing
either one will automatically help the other.
The dip in the lateral chart at 11.4 is caused mostly by
over braking at 10.6. This results in the small foot fake
at 11.2. Double braking hurts aggressiveness by "softening"
the average slope of the line. One way to think of this
aspect of aggressiveness is the "steepness" of the
transitions on both lateral and longitudinal charts.
Simultaneously, the over-braking makes it very difficult to
stay at the edge of the friction circle, and causes a
"bounce" in the Total G's chart. This unrealistic peak is
caused by the "snap" of heavily loading the car in an
unsustainable way. Smoothness suffers because this high
total g peak can't be sustained throughout the entire
corner.
So in that particular turn, you could focus on slightly more
smoothness, and probably gain aggressiveness automatically.
And so it is with all foot fakes.
These are very minor, but look at 17.3, 22.0, and 29.9.
These last two are gas foot fakes, which are even more
costly to your time. All ding both smoothness and
aggressiveness.
Another little curiosity is the extra hand movement at
24.8. Since I wasn't there I don't know if the course
required you to turn around a cone there, or if a better
line would have prevented this motion entirely. As we all
know, cars can go faster when they are pointed straight
ahead. Slip angle, even just the minimal slip angle at the
tread that is the initial component of steering at a minimal
1/2g level, causes friction that results in a loss of
forward acceleration.
And the biggest loss in Usage score on the entire run is
also interesting. It's the 1-2 shift at 2.6. We all tend
to ignore the importance of shifting, and how much that
coasting while the clutch is in costs us. Not that Todd's
0.7 second shift is all that bad, but once you start driving
at this level, every possible tenth gain matters.
So in summary, if you find yourself getting ratings that are
just "too high", (after you finish patting yourself on the
back, which you certainly deserve), reset your Sustain
Duration, or eliminate it entirely. The new ratings you see
will give you more to work on again to help you find those
last few tenths.
The vast majority of drivers out there won't need to do
this. The factory defaults of using Sustained Peaks for the
limits, with a 1/2 second sustain will work better for most
novice to intermediate drivers. And the factory defaults
are calibrated to what most of us need to see and work on.
But once you advance enough, you'll want to begin using
Absolute Peaks.
Finally, look past the ratings, and into the formations
themselves for additional clues as to where you can improve
your time. GEEZ ratings represent only one way of looking
at your performance on track. I use ratings for spotting
major trends, but as you improve to the point Todd is at
now, you'll often have to look deeper than just the ratings
to find the best sources for improvement.
It's GREAT to see this kind of discussion on the GEEZ list!
And thanks Todd for including the address for your zip file
with your run in it!
--Byron
Todd Green wrote:
>
> >to a faster time. Bob is likely taking a better *line* than you are in some
> >places, and getting through certain parts of the course quicker than you
> >even though you're pulling higher "numbers" in those areas.
>
> Bob and I walked the course together about 10 times. We shared where
> each of us were running and we were pretty much running the exact same
> line (we also talked about car attitude at key points, where we were
> getting on the throttle, etc. etc.) (From externally watching him, he
> also seemed to be running very much the same line as me.) Now if you
> are talking about who executed the plan better....well the times answer
> that story ;) That is not to say that there may have been subtle
> differences in the lines that accounted for the times.
>
> >If I understand how Geez figures this (Byron or someone correct me if I'm
> >wrong), the usage numbers are based on how closely the driver comes to
> >the peak numbers seen when he drives the car.
>
> This is how GEEZ works, but it does smoothness and agressiveness
> separately. Agressiveness is only measured when you *don't* have large
> lateral G's on the car. According to GEEZ I didn't have any major
> problems pointing the car straight and getting on the power (i.e. no
> happy feet or hands.) Where I did have the largest room for improvement
> was maintaining the lateral limits (again according to the rating
> system.)
>
> >Therefore lower "peaks" by another driver can actually generate usage
> >percentages similar to those of the first driver, as long as the second
> >driver stays equally close to his lower "peak".
>
> Agreed. My point was that this isn't necessarily a good thing, and that
> if you took the first driver's values and used them for the rating
> system of the second driver, the second driver's ratings would go down.
>
> But I get your point, that I should not necessarily be looking to
> improve my arcs, but pick a different line so as to minimize them and
> lengthen the "straights".
>
> >OK, a friendly wager: I'd bet that if Bob and you had both driven your car
>at
> >Lemoore, Geez would likely have shown your peak numbers as being *higher*
> >in some places on the course than his, but he actually may have negotiated
> >those portions of the course as quickly, or more so, than you.
>
> Good point. And this is precisely why I think GEEZ needs the ability to
> directly overlay runs (hint hint Byron ;) ;) ;) When runs start differing
> by 10ths or 100ths, it is pretty hard to eyeball the differences in
> separate windows, IMHO. For now, I just run them through a Perl script
> and graph them in Excel.
>
> Thanks for the insights!
> Todd
|